Loading...
Hydraulic fracture monitoring and characterization using low-frequency distributed acoustic sensing and machine learning
Zhu, Xiaoyu
Zhu, Xiaoyu
Citations
Altmetric:
Advisor
Editor
Date
Date Issued
2022
Date Submitted
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
2023-05-04
Abstract
This thesis discusses the mechanisms and applications of fiber optic sensing techniques, with a focus on the low-frequency distributed acoustic sensing (LF-DAS). We also present two studies: (1) a Chalk Bluff study to evaluate hydraulic fracture completion; and (2) a Hydraulic Fracture Test Site 2 (HFTS-2) study used LF-DAS to provide child-to-parent well fracture-hit surveillance using machine learning algorithms.
The first study uses low-frequency DAS measurements to monitor the hydraulic fracturing in the Chalk Bluff field in the Denver-Julesburg (DJ) Basin, Colorado. This is a low porosity, low permeability unconventional reservoir, and therefore, hydraulic fractures is essential for production and hydraulic fracturing completion evaluation is critical. By interpreting fracture-hits from cross-well LF-DAS data, we observed significant differences in hydraulic fracture propagation between the two targeted formations: Niobrara and Codell. The half length of hydraulic fractures in the Codell formation is more than twice as long as that in the Niobrara. In addition, hydraulic fractures propagate significantly faster in Codell than in Niobrara under the same pumping rate. The fracture orientations interpreted from the LF-DAS data also show fracture azimuth influenced by the adjacent well completions. Between the two formations, there are also strong cross-formation fracture connections, with different up-going and down-going fracture propagation velocities.
For the second case study, we developed an automated workflow for data processing and classification of LF-DAS signals. In an infill drilling scenario, fracture hits from infill well completion can cause damage to parent wells and lead to potential production losses. However, fiber equipped at the parent wells can measure strain perturbation induced by hydraulic fracture propagation at adjacent child wells. This workflow will detect fracture-hit precursors and warn the engineers to take actions to prevent the fracture hit. To detect fracture hit precursors ahead of time from the fracture hit, we used three machine learning methods: random forest, neural networks and bagging support vector machine classifier. Among the three tested methods, the random forest classifier achieved the best performance. We then applied the workflow to a case study using the Hydraulic Fracture Test Site 2 (HFTS-2) dataset and achieved 93% accuracy on fracture hit detection. With the random forest model trained on the HFTS-2 dataset, we also tested the model transferability on the Chalk Bluff dataset. It provides a valuable example of an automated fracture-hit early-warning system, and provides a versatile labelled dataset for future LF-DAS fracture hit research.
Associated Publications
Rights
Copyright of the original work is retained by the author.